我使用熊猫作为数据库替代品，因为我有多个数据库(Oracle, SQL Server等)，我无法使一个SQL等量命令序列。
我有一个表加载在一个DataFrame与一些列:
YEARMONTH, CLIENTCODE, SIZE, etc., etc.
在SQL中，计算每年不同客户端的数量将是:
SELECT count(distinct CLIENTCODE) FROM table GROUP BY YEARMONTH;
结果就是
201301 5000
201302 13245
我如何在熊猫中做到这一点?
当前回答
这是另一种简单得多的方法。假设你的数据框架名为daat，列名为YEARMONTH:
daat.YEARMONTH.value_counts()
其他回答
我也使用nunique，但如果你必须使用'min'， 'max'， 'count'或'mean'等聚合函数，这将是非常有用的。
df.groupby('YEARMONTH')['CLIENTCODE'].transform('nunique') #count(distinct)
df.groupby('YEARMONTH')['CLIENTCODE'].transform('min') #min
df.groupby('YEARMONTH')['CLIENTCODE'].transform('max') #max
df.groupby('YEARMONTH')['CLIENTCODE'].transform('mean') #average
df.groupby('YEARMONTH')['CLIENTCODE'].transform('count') #count
使用新的Pandas版本，可以很容易地获得数据帧:
unique_count = pd.groupby(['YEARMONTH'], as_index=False).agg(uniq_CLIENTCODE=('CLIENTCODE', pd.Series.count))
使用crosstab，这将返回比groupby nunique更多的信息:
pd.crosstab(df.YEARMONTH,df.CLIENTCODE)
Out[196]:
CLIENTCODE 1 2 3
YEARMONTH
201301 2 1 0
201302 1 2 1
稍加修改后，得到如下结果:
pd.crosstab(df.YEARMONTH,df.CLIENTCODE).ne(0).sum(1)
Out[197]:
YEARMONTH
201301 2
201302 3
dtype: int64
To obtain the number of different clients and sizes per year (i.e. number of unique values of multiple columns), use a list of columns: df.groupby('YEARMONTH')[['CLIENTCODE', 'SIZE']].nunique() Actually, the result from the above code can be obtained using SQL syntax on df using pandasql (a module built on pandas that lets you query pandas DataFrames using SQL syntax). #! pip install pandasql from pandasql import sqldf sqldf(""" SELECT COUNT(DISTINCT CLIENTCODE), COUNT(DISTINCT SIZE) FROM df GROUP BY YEARMONTH """) If you want to keep YEARMONTH as a column, i.e. the analog of the following SQL query SELECT YEARMONTH, COUNT(DISTINCT CLIENTCODE), COUNT(DISTINCT SIZE) FROM df GROUP BY YEARMONTH in pandas is the following (set as_index to False): df.groupby('YEARMONTH', as_index=False)[['CLIENTCODE', 'SIZE']].nunique() If you need to set custom names to the aggregated columns, i.e. the analog of the following SQL query: SELECT YEARMONTH, COUNT(DISTINCT CLIENTCODE) AS `No. clients`, COUNT(DISTINCT SIZE) AS `No. size` FROM df GROUP BY YEARMONTH use named aggregation in pandas: ( df.groupby('YEARMONTH', as_index=False) .agg(**{'No. clients':('CLIENTCODE', 'nunique'), 'No. size':('SIZE', 'nunique')}) )
不同的列以及其他列上的聚合
要获得任何列(在您的例子中是CLIENTCODE)的不同值数量，我们可以使用nunique。我们可以在agg函数中传递输入作为字典，以及其他列上的聚合:
grp_df = df.groupby('YEARMONTH').agg({'CLIENTCODE': ['nunique'],
'other_col_1': ['sum', 'count']})
# to flatten the multi-level columns
grp_df.columns = ["_".join(col).strip() for col in grp_df.columns.values]
# if you wish to reset the index
grp_df.reset_index(inplace=True)
